Table 2.

Performance metrics of the onetime fatigue detection approach.

Set up	System	AUCa (%)	P value	Specificity (%)	Sensitivity (%)	Precision (%)	F1-score (%)
150 keys	Fatigue (DMLb)	72.1	<.001	73	69	67	72.2
150 keys	Random forest	68.4	<.001	68	63	64.6	70.3
150 keys	Support vector machine	58.5	<.001	58	58	57.9	65.2
150 keys	k-nearest neighbor	58	<.001	77	51	64.6	70.3
150 keys	Fatigue (Softmax)	51.9	<.001	50	52	48	49.1
5 minutes	Fatigue (DML)	72.1	<.001	73	69	67	72.2
5 minutes	Random forest	77.8	<.001	70	76	66.3	71
5 minutes	Support vector machine	74.4	<.001	70	73	65.9	70.7
5 minutes	k-nearest neighbor	71.7	<.001	76	65	64.7	67.6
5 minutes	Fatigue (Softmax)	51.9	<.001	50	52	48	49.1

^aAUC: area under the curve.

^bDML: distance metric learning.